Michiko Watanabe

Intelligent AGV driving toward an autonomous decentralized manufacturing system

Abstract This paper proposes two methods that give intelligence to automatically guided vehicles (AGVs). In order to drive AGVs autonomously, two types of problems need to be overcome. They are the AGV navigation problem and collision avoidance problem. The first problem has been well known since 1980s. A new method based on the feature scene recognition and acquisition is proposed. The sparse distributed memory neural network (SDM) is employed for the scene recognition and acquisition. The navigation route for the AGV is learnt by use of Q-learning depending on the recognized and acquired scenes. The second problem is described as mutual understanding of behaviors between AGVs. The method of mutual understanding is proposed by the use of Q-learning. Those two methods are combined together for driving plural AGVs autonomously to deliver raw materials between machine tools in a factory. They are incorporated into the AGVs as the machine intelligence. In experimental simulations, it is verified that the first proposed method can guide the AGV to the suitable navigation and that the second method can acquire knowledge of mutual understanding of the AGVs’ behaviors.

GA applications to physical distribution scheduling problem

A physical distribution system has a number of optimization problems. Most of them belong to a combinatorial problem, to which conventional mathematical programming methods may hardly be applied. This paper reports on two applications of the genetic algorithm (GA) to physical distribution scheduling problems, which arise at real physical distribution centers. The developed GA schedulers took the place of conventional schedulers, which were coded by rule-based technologies. Advantages of the introduction of GA schedulers into the physical distribution system are as follows: (1) the GA becomes a general problem-solver engine. Once we develop this engine, we only have to develop interfaces for the applications; and (2) fitness functions necessary for the GA force the physical distribution schedulers to have approximate performance estimation. This was not taken into consideration when the rule-based scheduler was used. Two applications of the discussed schedulers were implemented with real distribution centers, and they brought much efficiency to their management.

Macroscopic quantitative observation of multi-robot behavior

It is very difficult to estimate behaviors of multiple autonomous robots or mutual interactions of them in real time. Therefore, we propose a quantitative observation approach of multiple robots behaviors. This approach introduces thermodynamic macroscopic state values to the multi-robot systems. The advantage of this approach is that it enables to observe the behaviors of autonomous robots in real world and can be mapped to characteristic values in another conceptual state space. Thermodynamic macroscopic state values, such as temperature, pressure and entropy, are defined in mobile robots systems. In our definition, each mobile robot is supposed to have a particle in thermodynamic systems. The experiment shows that the states of robots system can be classified by thermodynamic macroscopic state value. This verifies that the macroscopic quantitative observation is efficient and applicable to control multi-robot systems.

AGV autonomous driving based on scene recognition acquired by simplified SDM

An intelligent material handling system plays a great role in an autonomous decentralized manufacturing system (ADMS). An automatically guided vehicle (AGV) is at the center of the intelligent material handling system. This paper reports on a method for autonomously driving the AGV in the ADMS. A new method is proposed that combines the sparse distributed memory neural network (SDM) with Q-learning (Q-L). The SDM is adopted to explore and acquire scenes required for AGV driving. Q-L is employed to find a direction at the scene acquired by SDM. Numerical simulations verify that the SDM can extract the feature scenes necessary to drive the AGV and that Q-L instructs the suitable direction to the AGV at the extracted scenes towards the target location through its driving experiences.

An Artificial Neural Network Based on the Architecture of the Cerebellum for Behavior Learning

In the last decade, artificial intelligence (AI) pervades every aspect of our lives. However, there is a gap between AI-based machine behavior and human in natural communication. The behavior of most AI is determined as a task list generated by engineers, but to obtain high-level intelligence, AI needs the ability to cluster tasks from circumstances and learn a strategy for achieving each task. In this study, we focus on the human brain architecture that gives it the ability to self-organize and generalize sensory information. We propose an Artificial Neural Network (ANN) model based on that architecture. We describe a cerebellum-based ANN model (C-ANN) and verify its capacity to learn from the phototaxic behavior acquisition of a simple two-wheeled robot. As a result, the controller of the robot is self-organized to be simple and able to achieve positive phototaxis. This result suggests that the proposed C-ANN model has the capability of supervised learning.

MACROSCOPIC QUANTITATIVE OBSERVATION OF MULTI-ROBOT BEHAVIOR

In observing behavior of multiple autonomous robots, a microscopic observation expressed by dynamic equations is usually used. However, it is very difficult to estimate behavior of robots or mutual interactions among them in real time. Furthermore, it is hard to realize the observed system by taking consideration in all the factors of the system. On the other hand, a macroscopic observation defined by state equations is efficient for recognizing behavior of multiple robots. In this study, a quantitative observation approach is proposed on behavior of multiple robots. This approach introduces macroscopic state quantities in thermodynamics into expression of the multiple robots system. The advantage of this approach is that observation on the behavior of autonomous robots in real world can be mapped to characteristic quantities in another conceptual state space. At first, the state quantities of multiple robots system are defined and their physical meaning is discussed regarding it as quantitative observation of multiple robots. The macroscopic state quantities in thermodynamics, such as temperature, pressure and entropy, are introduced into mobile robots system. Each mobile robot is regarded as a particle in term of thermodynamic systems. Experiments show that the states of robots system can be classified from a viewpoint of the macroscopic state quantities in thermodynamics. This verifies that the macroscopic quantitative observation is efficient and applicable to controlling multiple robots system.

Multi-Vehicles Learning for Pushing a Bar-Shaped Object

Abstract This paper presents a method of multi-vehicles learning for pushing a barshaped object with their cooperation. Multi-vehicles are supposed to drive themselves without any human instruction and to convey the bar-shaped object to a given location by pushing it. The problem is to acquire autonomous and cooperative activities of the vehicles for achieving the above-mentioned purpose by trial and error. Q-learning is adopted to investigate emergence of such activities through trial and error experience. For a learning experiment, a simple simulation model is proposed by using the Newtonian physics for pushing the bar-shaped object by three vehicles. Numerical experiment shows that three vehicles can acquire the knowledge to successfully push the bar-shape object to the given goal in cooperation without any differential equations.

Autonomous Driving of Multiple AGVs for FMS by Use of SLA.

In order to efficiently operate FMS with the AGVs driving, it becomes important to control AGVs driving. This paper represents a method of autonomous driving for multiple AGVs equipped with FMS. It is assumed that an FMS model has a one-way driving lane for AGVs and neither passing nor encounter between AGVs is allowed. AGV is modeled by SLA (Stochastic Learning Automaton) to have a capability of making driving decision by itself. Such a capability is accomplished by the stochastic learning process of SLA. Numerical experiments show that AGV can learn proper driving actions and operate its run in the shortest possible distance.

A Study on Artificial Neural Network Structures for Agent Learning

Artificial neural networks which have been used for agent learning have mostly employed back-propagation and recurrent neural networks. We, however, have observed that there exists another network structure in life—a small world network, which is used by C-elegance, a kind of eelworms. We examined not only the performance of the small world network but that of a regular graph network and a random graph network. We applied these three networks to agent learning problems, and when we compared them with back-propagation and recurrent neural networks, it became clear that in the case of small world network structures, it has the same or even better performance as compared to back-propagation and recurrent neural networks despite a lower number of synapses.

Development of a GA-based packing planning system for on-site logistics

A recent tendency has been to suppress excessive capital investment in logistics centers. There have been many cases where computer-based logistics systems have been introduced in parallel with new equipment to handle the control of such equipment, as well as the unified processing of shipping, loading, unloading, and inventory management, etc. However, capital investment has recently been minimized and the introduction of computer-based management systems which retain tasks that may be achieved using manpower has not been unusual. The requirements on such systems to control equipment optimally are reduced, but the need has arisen for computer-based systems to assist the smooth operation of tasks performed using manpower. In this research, genetic algorithms (GAs) are applied to the design of plans for shipment packing, loading and dispatch occurring at logistics centers. A system supporting decision-making by workers through the design and presentation of efficient plans was formulated and implemented.